arXiv subjects

He Wang

Publications and source records attributed to He Wang.

At least 55 records · Page 3Linked to original sources

arXiv2026

PACE: Adaptive Budget Allocation for Time-Efficient Embodied Planning

Reasoning-enhanced large language models have achieved remarkable improvements in planning tasks, yet their deployment in embodied systems remains impractical due to prohibitive inference delays-often exceeding minutes per planning instance. The fundamental bottleneck stems from the serial nature of existing paradigms: models must complete all reasoning before any action execution, leaving execution time windows entirely unexploited. We introduce PACE (Planning with Adaptive Cognitive Effort), a framework that enables interleaved reasoning and execution through two key innovations: an Interleaved Think-Act architecture that pipelines cognitive processing with action execution, and a Dynamic Budget Allocator that adapts reasoning token budgets to available execution time windows. On the Robotouille benchmark using Qwen3-8B-AWQ, PACE achieves a 10% success rate-representing a 67% improvement over the ReAct+Think baseline-while delivering 6.9 times acceleration in thinking time compared to unconstrained reasoning. The framework hides 66.8% of thinking time within execution windows, demonstrating that strategic cognitive effort allocation can simultaneously improve both planning quality and time efficiency. These results provide evidence that time-aware architectural innovations enable reasoning models to operate in latency-sensitive embodied domains where they were previously impractical.

cs.RO↗

arXiv2026

Dynamic Pricing and Matching for Two-Sided Queues

Motivated by applications from gig economy and online marketplaces, we study a two-sided queueing system under joint pricing and matching controls. The queueing system is modeled by a bipartite graph, where the vertices represent customer or server types and the edges represent compatible customer-server pairs. Both customers and servers sequentially arrive to the system and join separate queues according to their types. The arrival rates of different types depend on the prices set by the system operator and the expected waiting time. At any point in time, the system operator can choose certain customers to match with compatible servers. The objective is to maximize the long-run average profit for the system. We first propose a fluid approximation based pricing and max-weight matching policy, which achieves an $O(\sqrtη)$ optimality rate when all the arrival rates are scaled by $η$. We further show that a two-price and max-weight matching policy achieves an improved $O(η^{1/3})$ optimality rate. Under a broad class of pricing policies, we prove that any matching policy has an optimality rate that is lower bounded by $Ω(η^{1/3})$. Thus, the latter policy achieves the optimal rate with respect to $η$. We also demonstrate the advantage of max-weight matching with respect to the number of server and customer types $n$. Under a complete resource pooling condition, we show that max-weight matching achieves $O(\sqrt{n})$ and $O(n^{1/3})$ optimality rates for static and two-price policies, respectively, and the latter matches the lower bound $Ω(n^{1/3})$. In comparison, the randomized matching policy may have an $Ω(n)$ optimality rate.

math.OC↗

arXiv2026

A Cooperative Implementation of Mesh Stability in Vehicular Platoons

This work studies the problem of mesh stability in connected and automated vehicles. Mesh stability, also known as 2D string stability, refers to studying how disturbances propagate in vehicular platoons in both longitudinal and lateral direction. As opposed to available decentralized results only relying on on-board sensing, the distinguishing feature of this work is a cooperative version of mesh stability, where onboard sensing is augmented by vehicle-to-vehicle communication. This cooperative version dramatically improves the state-of-the-art decentralized performance: for longitudinal control, a new cooperative non-identical protocol (i.e. with non-identical control gains) is proposed that improves the state-of-the-art decentralized non-identical protocol in terms of scalability of the control gains and strong notion of string stability. For lateral control, after showing that the non-identical approach is not necessary, a cooperative non-identical protocol is proposed to achieve another strong notion of string stability. Robustness of the proposed implementation against vehicle-to-vehicle communication delays and actuation time lags is considered. Numerical experiments, also performed with the vehicle simulator CarSim, validate the robustness and effectiveness of the proposed protocol.

eess.SY↗

arXiv2026

GraspADMM: Improving Dexterous Grasp Synthesis via ADMM Optimization

Synthesizing high-quality dexterous grasps is a fundamental challenge in robot manipulation, requiring adherence to diversity, kinematic feasibility (valid hand-object contact without penetration), and dynamic stability (secure multi-contact forces). The recent framework Dexonomy successfully ensures broad grasp diversity through dense sampling and improves kinematic feasibility via a simulator-based refinement method that excels at resolving exact collisions. However, its reliance on fixed contact points restricts the hand's reachability and prevents the optimization of grasp metrics for dynamic stability. Conversely, purely gradient-based optimizers can maximize dynamic stability but rely on simplified contact approximations that inevitably cause physical penetrations. To bridge this gap, we propose GraspADMM, a novel grasp synthesis framework that preserves sampling-based diversity while improving kinematic feasibility and dynamic stability. By formulating the refinement stage using the Alternating Direction Method of Multipliers (ADMM), we decouple the target contact points on the object from the actual contact locations on the hand. This decomposition allows the pipeline to alternate between updating the target object points to directly maximize dynamic grasp metrics, and adjusting the hand pose to physically reach these targets while strictly respecting collision boundaries. Extensive experiments demonstrate that GraspADMM significantly outperforms state-of-the-art baselines, achieving a nearly 15\% absolute improvement in grasp success rate for type-unaware synthesis and roughly a 100\% relative improvement in type-aware synthesis. Furthermore, our approach maintains robust, physically plausible grasp generation even under extreme low-friction conditions.

cs.RO↗

arXiv2026

Scaling Behavior Foundation Model for Humanoid Robots

Humanoid control requires natural whole-body coordination, precise real-time responses to control signals, and robust generalization across diverse environmental contexts, making it a cornerstone for generalist embodied agents. Behavior Foundation Models (BFMs) have recently emerged as a promising solution to address these challenges by leveraging large-scale behavioral data to achieve superior expressiveness, versatility and generalization. However, despite growing interest in scaling BFMs to further improve their capabilities, it remains unclear how key factors, including the learning paradigm, behavioral data and model architecture should be coordinated to enable effective scaling. In this work, we revisit the scaling recipe for BFMs and demonstrate that substantial performance gains can be achieved through the coordination of three core components: 1) the learning paradigm of motion tracking that reformulates diverse humanoid control problems as the reproduction of integrated whole-body behaviors in the global frame; 2) the strategic synergy between on-policy rollout quantity and reference motion diversity; and 3) the expressive and scalable model architecture termed Humanoid Transformer that facilitates the natural emergence of structured behavioral representations. Through extensive experiments in both simulation and real-world deployment, we demonstrate that our approach yields significant improvements in control fidelity and task generalization, reducing Mean Per-Keypoint Position Error (MPKPE) on the test set by over 10% in local mode and 82% in global mode compared with existing humanoid controllers. These results establish BFM as a principled and effective foundation for scalable and general-purpose humanoid control.

cs.RO↗

arXiv2026

WAM-TTT: Steering World-Action Models by Watching Human Play at Test Time

Steering robot foundation models (RFMs) toward new task variants or user-preferred behaviors remains challenging, often requiring additional robot demonstrations, task-specific fine-tuning, or long-context conditioning. We present WAM-TTT, a test-time training framework for steering world action models from raw human videos. Rather than treating human videos as trajectories to imitate, WAM-TTT absorbs them into a lightweight adaptive memory inside a frozen WAM through self-supervised video prediction. To make this memory useful for control, we introduce a meta-training stage that aligns human demonstrations with robot behaviors using paired human-robot data and a key--value memory reconstruction objective. At test time, only unlabeled human videos are required to adapt the memory, while the pretrained WAM remains frozen. This enables efficient and reusable steering without robot actions, human-side annotations, or task-specific fine-tuning, while preserving the generalization ability of the foundation model. Extensive experiments show that WAM-TTT consistently outperforms in-context human-video conditioning baselines across diverse manipulation tasks and generalization settings.

cs.RO↗

arXiv2026

Fictional Worldbuilding: Multi-Agent LLM Collaboration with Hierarchical Context Compression and Iterative Review

Worldbuilding, the construction of coherent fictional worlds, is a foundational task in game design and literary creation. Large Language Models (LLMs) offer new possibilities for automated content generation, but their application to worldbuilding faces three challenges: context explosion that grows linearly with the building process, the tension between creative diversity and content consistency, and the absence of automated quality assurance. This paper presents AutoWorldBuilder, a multi-agent collaborative system that addresses these challenges through five integrated components: a structured concept network with conflict detection; a DAG-based hybrid batch scheduler that groups tasks by semantic locality; a four-layer context compression mechanism achieving approximately 90% token reduction; an iterative review system with specialized Auditor agents that improves proposal pass rates from 42% to over 85%; and a skill-driven agent architecture supporting zero-code extension with differentiated temperature configuration. Two experiments across 20 diverse worldbuilding tasks, using GPT-OSS 120B and DeepSeek v3.2 as LLM backends, demonstrate a 95.0% success rate. The system generated 56-103 self-consistent concepts per world in 18-31 minutes with zero-conflict delivery. The architectural patterns validated here, including layer-as-budget compression, semantic-locality scheduling, and separation of generation and review, transfer to the broader class of knowledge-intensive, multi-agent LLM applications.

cs.AI↗

arXiv2026

EdgeBench: Unveiling Scaling Laws of Learning from Real-World Environments

Pretraining scaling laws reveal that model capability improves predictably with data and compute. But learning from real world environments after deployment remains far less understood. Analyzing roughly 38,000 hours of agent interaction with the environment across 134 real world tasks, we find, to the best of our knowledge, the first evidence that overall performance during environment learning follows a log-sigmoid scaling law with remarkably high precision, reaching R^2 = 0.998. Across model generations, we also find that agent learning speed roughly doubles every three months. This discovery stems from EdgeBench, a suite of 134 real world tasks with ultra-long horizons, spanning scientific discovery, software engineering, combinatorial optimization, professional knowledge work, formal mathematics, and interactive games. Each task sustains at least 12 hours of continuous agent operation under rich, multilevel feedback, and is built through substantial expert effort. We publicly release 51 tasks and our full evaluation framework to accelerate the study of how agents learn from real world experience.

cs.CL↗

arXiv2026

Rigorous analysis of the time-splitting methods for the semiclassical Dirac equation

We provide rigorous error analysis of the mass-preserving time-splitting methods for solving the semiclassical Dirac equation. The scaled Planck constant $ε$ in the equation gives rise to rapid oscillations in both space and time when $0<ε\ll 1$ with wavelengths of order $O(ε)$. %We prove that the first-order splitting $S_1$ and the second-order splitting $S_2$ schemes preserve the total discretized mass. Rigorous error estimates reveal the precise dependence of the approximation errors on the time step $τ$, the spatial mesh size $h$, and the parameter $ε$. Specifically, the temporal error scales as $O\left(τ/ε^2\right)$ for the first-order splitting $S_1$ and as $O\left(τ^2/ε^3\right)$ for the second-order splitting $S_2$, while the spatial error scales as $O(h^m/ε^m)$ for both methods, where $m$ is related to the regularity of the solution. In addition, we obtain error bounds for key physical observables, including the total probability density $ρ$ and the current density $\mathbf{J}$. Compared with finite difference time domain (FDTD) methods, time-splitting approaches exhibit spectral accuracy in space and retain a relatively low computational cost. Furthermore, we demonstrate that higher accuracy can be achieved by employing the fourth-order compact time-splitting ($S_\text{4c}$) method. Numerical experiments are conducted to verify the reliability of the error estimates.

math.NA↗

arXiv2026

MM-Nav: Multi-View VLA Model for Robust Visual Navigation via Multi-Expert Learning

Visual navigation policy is widely regarded as a promising direction, as it mimics humans by using egocentric visual observations for navigation. However, optical information of visual observations is difficult to be explicitly modeled like LiDAR point clouds or depth maps, which subsequently requires intelligent models and large-scale data. To this end, we propose to leverage the intelligence of the Vision-Language-Action (VLA) model to learn diverse navigation capabilities from synthetic expert data in a teacher-student manner. Specifically, we implement the VLA model, MM-Nav, as a multi-view VLA (with 360 observations) based on pretrained large language models and visual foundation models. For large-scale navigation data, we collect expert data from three reinforcement learning (RL) experts trained with privileged depth information in three challenging tailor-made environments for different navigation capabilities: reaching, squeezing, and avoiding. We iteratively train our VLA model using data collected online from RL experts, where the training ratio is dynamically balanced based on performance on individual capabilities. Through extensive experiments in synthetic environments, we demonstrate that our model achieves strong generalization capability. Moreover, we find that our student VLA model outperforms the RL teachers, demonstrating the synergistic effect of integrating multiple capabilities. Extensive real-world experiments further confirm the effectiveness of our method.

cs.RO↗

arXiv2026

Self-supervised Garment Dynamics with Persistent Wrinkles

Self-supervised neural garment simulation has become popular due to its computational efficiency, good visual realism, and no reliance on training data. However, existing methods greatly simplify the mechanical properties of fabrics, ignoring persistent wrinkles caused by plasticity. Although this simplification allows for modeling of purely elastic material and simple training via energy minimization, the lack of believable wrinkles adversely affects the visual realism. Therefore, we introduce the first self-supervised neural garment simulator that explicitly models persistent wrinkles. This is accomplished through a novel physics-inspired loss function, which turns learning into a moving energy minimization problem to mimic plasticity. However, this requires learning to use a changing loss function, which causes difficulties in training because the loss function changes during optimization. To this end, we propose a new physics-inspired curriculum learning scheme where the target material for learning gradually changes from pure elasticity to elasto-plasticity, allowing the loss function and the learnable parameters to jointly converge. Through a comprehensive evaluation, we show that for the first time, self-supervised learning models can generate natural persistent wrinkles, outperforming existing methods on a variety of garments, body shapes, and body motions, according to a range of metrics.

cs.GR↗

arXiv2026

StereoVLA: Enhancing Vision-Language-Action Models with Stereo Vision

While Vision-Language-Action (VLA) models excel in generalist manipulation, they often lack fine-grained spatial awareness and show limited viewpoint robustness. This limitation largely stems from the reliance on pretrained RGB encoders, which lack explicit geometric cues and prioritize semantic alignment over geometric representation. We argue that effective visual representations for VLA models must jointly encode both semantic and geometric information. In this paper, we introduce StereoVLA, the first VLA model to incorporate rich geometric cues from large-scale synthetic stereo data. StereoVLA employs a Geometric-and-Semantic (GeoSem) vision encoder that extracts geometric cues from subtle stereo-view disparities for precise spatial perception, while simultaneously capturing semantic features from pixel observations to support language-conditioned manipulation. Additionally, we introduce two synergistic co-training objectives: Interaction-Region Depth Estimation for precise spatial reasoning, and Camera Parameter Estimation to implicitly align perception and action coordinate systems. Compared with baselines that employ various input modalities, StereoVLA achieves a 33.4% absolute gain in success rate in real-world experiments and demonstrates robustness to near-hemispheric camera perspectives. Project page: https://shengliangd.github.io/StereoVLA-Webpage.

cs.RO↗

arXiv2026

GAPartManip: A Large-scale Part-centric Dataset for Material-Agnostic Articulated Object Manipulation

Effectively manipulating articulated objects in household scenarios is a crucial step toward achieving general embodied artificial intelligence. Mainstream research in 3D vision has primarily focused on manipulation through depth perception and pose detection. However, in real-world environments, these methods often face challenges due to imperfect depth perception, such as with transparent lids and reflective handles. Moreover, they generally lack the diversity in part-based interactions required for flexible and adaptable manipulation. To address these challenges, we introduced a large-scale part-centric dataset for articulated object manipulation that features both photo-realistic material randomization and detailed annotations of part-oriented, scene-level actionable interaction poses. We evaluated the effectiveness of our dataset by integrating it with several state-of-the-art methods for depth estimation and interaction pose prediction. Additionally, we proposed a novel modular framework that delivers superior and robust performance for generalizable articulated object manipulation. Our extensive experiments demonstrate that our dataset significantly improves the performance of depth perception and actionable interaction pose prediction in both simulation and real-world scenarios. More information and demos can be found at: https://pku-epic.github.io/GAPartManip/.

cs.RO↗

arXiv2026

CLAR: Learning 3D Representations for Robotic Manipulation by Fusing Masked Reconstruction with Multi-Level Contrastive Alignment

The spatial information inherent in 3D point clouds is crucial for robotic manipulation. However, existing 3D pre-training methods face a fundamental trade-off: Masked Autoencoding (MAE) excels at capturing spatial-geometric features but lacks semantics, whereas contrastive learning, while able to distill semantics from 2D foundation models, is ill-suited for the fine-grained details required for manipulation tasks. To address these challenges, we propose CLAR, a novel 3D pre-training framework that synergizes global understanding with fine-grained local alignment. Our framework unifies MAE with global cross-modal contrastive learning to integrate robust spatial awareness with rich semantic understanding. To enhance its focus on fine-grained details, at the local level, we introduce an adaptive alignment mechanism that leverages deformable attention to force precise correspondences between local 3D geometry and 2D visual features, thereby overcoming the limitations of conventional global alignment in manipulation tasks. Extensive experiments in simulation and the real world demonstrate that CLAR achieves state-of-the-art performance, significantly outperforming existing methods in visuomotor policy learning.

cs.RO↗

arXiv2026

AllDayNav: Lifelong Navigation via Real-World Reinforcement Learning

Lifelong embodied navigation in dynamic environments requires robots to form persistent scene understanding from fragmentary observations, which remains difficult for existing methods that rely on explicit maps or scene graphs and struggle to generalize beyond structured settings. We propose AllDayNav, a lifelong self-learning navigation framework that implicitly encodes scene dynamics into the billion-scale parameters of a large model via reinforcement learning, powered by a self-evolving multimodal memory that maintains and updates visual keyframes, semantic descriptions, and temporal context while autonomously generating open-vocabulary instructions, image goals, and structured rewards. Experiments in both synthetic and real-world environments across cross-room, cross-episode, and cross-task scenarios show that AllDayNav achieves success rates approaching $100\%$ and consistently surpasses strong map-based, VLM, and RL baselines in path efficiency and robustness, demonstrating implicit, memory-driven reinforcement learning as a scalable alternative to explicit mapping for reliable lifelong navigation.

cs.RO↗

arXiv2026

Graphical X Splatting (GraphiXS): A Graphical Model for 4D Gaussian Splatting under Uncertainty

We propose a new framework to systematically incorporate data uncertainty in Gaussian Splatting. Being the new paradigm of neural rendering, Gaussian Splatting has been investigated in many applications, with the main effort in extending its representation, improving its optimization process, and accelerating its speed. However, one orthogonal, much needed, but under-explored area is data uncertainty. In standard 4D Gaussian Splatting, data uncertainty can manifest as view sparsity, missing frames, camera asynchronization, etc. So far, there has been little research to holistically incorporating various types of data uncertainty under a single framework. To this end, we propose Graphical X Splatting, or GraphiXS, a new probabilistic framework that considers multiple types of data uncertainty, aiming for a fundamental augmentation of the current 4D Gaussian Splatting paradigm into a probabilistic setting. GraphiXS is general and can be instantiated with a range of primitives, e.g. Gaussians, Student's-t. Furthermore, GraphiXS can be used to `upgrade' existing methods to accommodate data uncertainty. Through exhaustive evaluation and comparison, we demonstrate that GraphiXS can systematically model various uncertainties in data, outperform existing methods in many settings where data are missing or polluted in space and time, and therefore is a major generalization of the current 4D Gaussian Splatting research.

cs.GR↗

arXiv2026

Unification of Closed-Open Industrial Detection Scenarios: New Large-Scale Benchmarks,Challenges and Baselines

Large-scale Visual-Language Models (LVLMs) have achieved remarkable success in natural visual tasks, yet their application to industrial defect detection remains challenging due to two fundamental limitations: (i) the scarcity of large-scale industrial datasets that cover diverse defect categories across multiple domains, and (ii) the reliance on manual prompts (points, boxes, masks) that introduce subjective noise and lack text-visual interaction for fine-grained understanding. To address these challenges, we introduce a Large-Scale Multi-Modal Industrial Open-Closed benchmark (MMIOC-1M) containing over one million samples across $14$ super-categories, $29$ industrial scenes, and $351$ defect subcategories. To our knowledge, MMIOC-1M is the first unified largest benchmark supporting both open-vocabulary and closed-set industrial detection, providing valuable pre-training data for LVLMs in industrial scenarios. Furthermore, we propose a Refined Text-Visual Prompt Network (RTVPNet) that incorporates three key innovations: (1) an expert-assisted domain projection mechanism that enables rapid adaptation of general vision models to industrial domains, (2) an energy-based sparse sampling strategy that automatically generates refined visual prompts without manual intervention, and (3) a bidirectional text-visual interaction module that enhances cross-modal semantic alignment and understanding. Extensive experiments demonstrate that RTVPNet achieves state-of-the-art performance on MMIOC-1M, LVIS, and COCO benchmarks while maintaining computational efficiency. The dataset and code are available at https://github.com/hellozzk/MMIO.

cs.AI↗

arXiv2026

LIMMT: Less is More for Motion Tracking

We argue that high-quality motion data can steer tracking policies toward better optimization trajectories early in training. In this work, we introduce LIMMT (Less Is More for Motion Tracking). To our knowledge, this is the first data-centric study for physics-based humanoid motion tracking. We go beyond simply removing low-quality and erroneous clips, but define motion data quality through three dimensions: physics feasibility, diversity, and complexity. We show that even training with under 3% of AMASS yields better tracking performance than training with the full dataset. We further conduct data cleaning on the estimated web-sourced mocap data. Extensive experiments and analyses validate the effectiveness of our framework.

cs.RO↗